Achieving precision in the language used is what the process is all about :)


The reason I think it is useful to make a distinction is that the value
came from identifying a security policy violation and pinning generally
understood to encompass enforcement of a security policy.

Getting a security policy mechanism to the point where it is enforceable is
hard. Google has asked us not to implement the pinning code in Comodo
Dragon even for google.com for operational reasons that look valid to me.


We can use Perspectives or Convergence or similar to detect and report
violations in cases where we do not have sufficiently good data to force
connection failure.

It is clearly desirable to have the client have the capability to fail a
connection but too many false positives will result in people turning the
systems off (unless they are in Iran).

I think we will see security policy coming in different degrees of quality:


0) No information

1) Weak - e.g. data that has been collected from passive observations and
heuristic means.

2) Good - the data comes from an authoritative source (e.g. HTTP headers,
DNS records)

3) Strong - there is authenticated, curated data from an authoritative
source backed up by data from a trusted source

4) Proof - the client has the benefit of a full cryptographic
authentication of the security policy that is tied to immutable sources
(e.g. Merkle tree based notary).


At the moment I see a lot of proposals that attempt to find the sweet spot
on this continuum. Well maybe that is the wrong approach for the protocol
to take. Maybe a better approach is to have a protocol that supports
transport of security policy data of any quality and allows the client to
make an appropriate decision. This would also take into account the degree
of concern the user had. Iranian dissidents and security specialists would
probably want to be more aggressive than regular users.

So a default approach would probably be for the client to merely report
policy violations against weak policy to some external party, warn the user
if there is a violation against good policy and block the connection on
Strong policy or Proof.


Getting to proof is good (yeah, I did the formal methods thing too). But it
is also very difficult to achieve.


On Fri, Jan 20, 2012 at 1:59 PM, Chris Palmer <[email protected]> wrote:

> On Fri, Jan 20, 2012 at 8:14 AM, Phillip Hallam-Baker <[email protected]>
> wrote:
>
> >> Without pinning, fraudulent certificates are invisible, especially when
> >> applied to specific individuals. The fraudulent diginotar cerrtificates
> in
> >> Iran would have gone unnoticed had it not been for the pinning of Google
> >> certificates.
> >
> > The Google certs were not pinned. Pinning is a very specific mechanism.
>
> We consider them to have been pinned. The dynamic pinning proposal in
> the WebSec WG is a generalization of what we call "preloaded" (or
> "static") pins.
>
> Word-noodling aside, the attack might very well have remained
> invisible without Chrome's knowledge of and insistence on Gmail's true
> SPKIs.
>
>
> --
> They who can give up general-purpose computing to obtain a little
> temporary safety, deserve neither general-purpose computing nor
> safety.
>



-- 
Website: http://hallambaker.com/
_______________________________________________
therightkey mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/therightkey

Reply via email to